Skip to main content
Claude, Cursor, and other clients call the Keldyn API through the Model Context Protocol. They can read frameworks and controls, record evidence, and update GRC records. Every request runs as you, on the team you pick at sign-in. Below is how to connect a client.
The server speaks Streamable HTTP and authenticates with OAuth 2.1. There is no API key. You sign in through the browser. That is the only way to connect.
Running a security review before you connect KC, or just want to know the ground rules?

Data security

Encryption, access, and how MCP access differs from your coding agent’s own access.

Cybersecurity

The security frameworks we align with and how the service is tested for vulnerabilities.

Compliance

Model training, data location, subprocessors, GDPR/DPA, attestations.

Responsible use

What KC is for, what it isn’t for, and where human judgement stays in charge.

Data flow

Where your data goes, step by step, from tool call to answer.

​
Find your MCP URL

The Keldyn MCP server answers every request at the root path (/):
Keep the trailing slash on the URL, and use the exact host your client connects to. The URL is the OAuth resource identifier, so it must match byte-for-byte.

​
Prerequisites

Before connecting, confirm the following:
  • Keldyn account: You sign in with your normal Keldyn credentials in the browser during the OAuth flow.
  • Team membership: You must belong to at least one team. You’ll pick a team during sign-in, and every request is scoped to it.
  • Team admin (for some tools): Listing integration-collected control evidence (list-control-evidence) and all Trust Center management tools (*-trust-center-*) require a team-admin (or owner) token.
  • Supported AI tool: Any client that supports remote MCP servers with OAuth, including Claude Code, Cursor, VS Code, and Codex.

​
Claude Code

1

Add the MCP server

Run the following command in your terminal:
2

Authenticate

In Claude Code, run /mcp and select keldyn. A webpage will open in your browser: sign in, choose the team you want to act as, and click Allow.
3

Start using Keldyn MCP

Once authorized, you’re ready to start querying and updating your Keldyn data from Claude Code.

​
Add to Cursor

1

Open Cursor settings

Open Settings in Cursor.
2

Add a custom MCP server

Select Tools & MCP → Add custom MCP.
3

Add the Keldyn configuration

Paste the following configuration:
4

Authenticate

Cursor opens the Keldyn consent page in your browser. Sign in, choose a team, and approve. Cursor completes the OAuth handshake automatically.

​
Other MCP-compatible clients

Keldyn MCP is a remote MCP server over Streamable HTTP. Any client that can do OAuth against a remote endpoint can try it, including VS Code (with an MCP extension), Codex, and newer clients. Claude Code and Cursor are the paths we actually test. Other clients usually work with the same two settings, and we have not formally validated them:
  • Server URL: https://mcp.keldyn.ai/. See Find your MCP URL.
  • Authentication: OAuth. Sign in as a Keldyn user and pick a team when the browser flow opens.
Your client’s docs say where a remote MCP server gets registered. Dynamic Client Registration and the OAuth handshake are automatic. The client reads /.well-known/oauth-protected-resource and opens the consent page.

​
How authentication works

Keldyn MCP is an OAuth 2.1 resource server. You never paste a token. The first connection walks discovery and consent:
1

Discovery

Your client makes an unauthenticated request, receives a 401 with a WWW-Authenticate challenge, and reads the server’s Protected Resource Metadata to discover Keldyn’s Authorization Server.
2

Registration & consent

Your client registers itself (Dynamic Client Registration) and opens the Keldyn consent page (PKCE, S256). You sign in, select a team, and approve.
3

Token exchange

Your client exchanges the authorization code for a short-lived access token (about 1 hour) plus a refresh token, and calls the MCP endpoint with Authorization: Bearer <token>.
Access tokens are scoped to the team you selected, so tools act as the signed-in user on that team. When a token expires, your client silently refreshes it and retries. To revoke access, sign out or revoke the connection from your Keldyn account.

​
What your AI agents can do with Keldyn MCP

Once connected, the assistant calls the Keldyn API as you. Organization context comes from the team you signed in with. You do not pass an org id on each call.

​
Frameworks & use cases

  • list-frameworks: Browse compliance/security frameworks (scope to a use case, the org, or all frameworks).
  • list-use-cases: List use cases for your team, with pagination and filters (search, status, risk level, business area, owner).
  • get-use-case: Fetch a single use case, including its frameworks and details.
  • update-use-case-architecture: Create or update a use case’s Threagile architecture model.
  • analyze-use-case-architecture: Run or fetch Threagile security analysis plus RCM architecture compliance checks for a use case, returning a unified, severity-sorted, paginated list of fixable issues.
  • update-risk-tracking: Record a disposition (mitigated, accepted, false-positive, in-progress, in-discussion) for a single Threagile security risk, with who/why.

​
Controls, tracking & evidence

  • list-controls: List RCM controls across use-case and org scopes, each with implementation status and a human-readable status label.
  • get-control: Fetch a single control by rcmId with all API fields plus its tracking entry (status, evidence text, assignee, finding, timestamps).
  • update-control: Update a control’s implementation tracking (todo, in_progress, complete, ignored) and optionally set evidence text, assignee, or a finding.
  • list-control-links: List audits, measures, obligations, or documents linked to a control.
  • list-control-evidence: List integration-collected (automated) control evidence with freshness/status. (Requires a team-admin token.)
  • list-evidences / get-evidence: Browse and fetch manual measure evidence (files and links).

​
Audits

  • list-audits / get-audit: List audits for the organization and fetch a single audit.
  • list-audit-controls: List the controls in scope for an audit.

​
GRC entities

  • list-entities / get-entity: Access risks, measures, obligations, findings, tasks, and data records.
  • list-entity-links: Traverse links between entities (e.g. risks → measures, obligations → controls).

​
Knowledgebase

These tools search the team’s Knowledgebase index: Library uploads, use case documents, and connected Drive, Confluence, Notion, and Otter sources. They require Read organisational context and a Keldyn-admin session, the same gates as the web app.
  • search-knowledge: Hybrid search over the index. Returns cited passages, not a drafted answer.
  • answer-knowledge: Same retrieval with neighbouring context and grounding instructions so your agent drafts a cited answer itself.
  • list-knowledge-tree / fetch-knowledge-document / fetch-document-section / fetch-knowledge-source: Browse the index as folders, open a file by path, read one section at a time, or inspect a connector’s sync status.
  • create-knowledge-document: Write a manual entry the same way New file does in the index browser.
The search-org-context prompt lists the team’s real sources up front so an agent can tell “not documented” from “connector never synced”, and instructs it not to answer from general knowledge.

​
Tasks

  • list-tasks / get-task: List and fetch organization tasks, filtered by measure, state, priority, or search.
  • create-task / update-task / delete-task: Full task CRUD; link a task to the control it addresses with rcmId.

​
Trust Center

These tools manage the organization’s public Trust Center. They require a team-admin token. The organization comes from the team on the token. How the public page works, and which image files it expects, is on Trust Center.
  • get-trust-center-status / update-trust-center-status: Check whether the public Trust Center is enabled (with its slug and public URL) and turn it on or off.
  • get-trust-center-profile / update-trust-center-profile: Read and update branding and contact settings (summary, primary color, contact email, privacy policy link, controls-updated badge). Image assets are uploaded from the web app.
  • list-trust-center-frameworks / update-trust-center-frameworks: List assigned frameworks with their publish settings and publish/unpublish badges, including planned/achieved status and target quarter/year.
  • list-trust-center-controls / set-trust-center-control-sharing: List your controls grouped by framework with their public sharing state, and show or hide an individual control on the public page.
  • list-trust-center-resources / update-trust-center-resource / delete-trust-center-resource: Manage the published document library (title, description, framework tag, label, published state, and public vs. access-gated visibility). New file uploads are done from the web app.
  • list-trust-center-domains: List configured custom domains and their verification status (read-only; domains are added and verified in the web app).
  • list-trust-center-access-requests / grant-trust-center-access-request / deny-trust-center-access-request: Review visitor requests for gated resources and approve or decline them. Approving emails the requester a 45-day access link; declining notifies them.
  • list-trust-center-access-grants / deactivate-trust-center-access-grant: List issued access links and revoke one immediately.
  • get-public-trust-center: Preview the full public payload (branding, published frameworks, shared controls, resources, subprocessors) for a slug, domain, or organization.

​
Built-in workflows (prompts)

The server also ships prompts. Each one loads your controls and pairs them with whatever the agent can see in the local repo:
  • review-changes: Review your code changes against the org’s Keldyn controls. The Keldyn Review Bot uses the same rules on open pull requests.
  • review-codebase: Review the whole codebase (or a scoped path/package) against controls.
  • start-dev: Prime the agent with developer guidance from applicable controls.
  • generate-architecture: Infer a Threagile architecture from your codebase and save it to a use case.
  • collect-evidence: Scan your codebase for positive evidence that code-related controls are satisfied, then confirm before saving.
  • remediate-insufficient-evidence: Find controls whose evidence failed acceptance criteria and gather stronger file-backed proof.
  • search-org-context: Search Knowledgebase with the team’s real sources listed up front, and refuse to answer from general knowledge.
Copy-paste prompts for kickoff, design, implement, review, evidence, and release are in the Developer SDLC playbook.

​
Resources

Model Context Protocol

Learn more about the open protocol that powers the Keldyn MCP server and other AI tool integrations.

MCP Inspector

Interactively test and debug an MCP server: list and call the Keldyn tools, prompts, and resources against a running instance.